Feature Flag & Experiment Platform · Documentation
Overview
What Feature Flag & Experiment Platform does in the ecosystem, how it connects, and what is simulated.
← Feature Flag & Experiment PlatformWhat it does
Feature Flag & Experiment Platform is the ecosystem's release-control plane. It manages feature flags — on/off, percentage rollout, and a targeting rule — and A/B experiments with real statistical significance, so the platform can ship behind a flag and decide with data instead of intuition.
OPERATE shows the flag board (live state, rollout %, targeting) and the experiment results: per-variant conversion, the lift, and a two-proportion z-test marked significant at the 95% level (|z| ≥ 1.96). CONFIGURE, which is owner-only, lets the owner enable a flag, move its rollout slider, set targeting, and ramp or roll back an experiment.
How it connects
The flags here are the ones gated apps read at assignment time — for example #04 Recommendation's ranker version and #10 Pricing's guardrail. Configuration changes are real events on the shared bus: flag.updated and experiment.ramped, both audited. It owns its flag and experiment config (demo_eco_c8_feature_flag, demo_eco_c8_experiment); edits are owner-only (C8 rejects viewer writes), persisted canonical and setWhere-guarded, while a viewer's view is read-only.
What is simulated, and the honesty note
This app is fully deterministic — there is no metered model stage, so it writes no cost-ledger row and shows no Cloud/OSS toggle; the significance figure is a closed-form z-test, not an estimate. Experiment exposures and conversions are synthetic, with a small deterministic lift baked in so significance is meaningful to demonstrate. Assignment is in-demo and labelled; in Stage-2 the same method surface maps to a real flag/assignment store with no UI change. No real users or PII appear anywhere.